Claude moves into Azure Foundry GA: first run of GB300 clusters
On June 29, Anthropic will officially make the Claude model family generally available (GA) on Microsoft Foundry on Azure. With the Opus 4.8 and Haiku 4.5 dual models, Azure customers can use them directly in Azure clients, with support for Azure Active Directory authentication, unified billing, and full integration with the Anthropic Messages API. This deployment is among the first frontier AI services running on NVIDIA GB300 Blackwell Ultra GPU clusters in Azure. With the launch of both Opus 4.8 and Haiku 4.5, pricing is the same as buying directly from Azure: Opus 4.8 input pricing is USD 15 per million tokens, and output pricing is USD 30 per million tokens; Haiku 4.5 input pricing is USD 0.25 per million tokens, and output pricing is USD 1.25 per million tokens. Azure customers can call the service through the Foundry unified interface without needing to apply for an Anthropic account or handle cross-platform billing. Azure AD authentication allows enterprise customers to use their existing identity management mechanisms to control access permissions for using Claude models.
First NVIDIA GB300 Blackwell Ultra deployment: Azure Accelerator upgrade
What’s special about this deployment is the underlying infrastructure: it runs entirely on NVIDIA’s next-generation GB300 Blackwell Ultra GPU clusters and is the first frontier AI service on Azure to enable this architecture. GB300 is NVIDIA’s latest AI accelerator announced in June; compared with the previous generation GB200, it delivers significant improvements in inference throughput and energy efficiency. This deployment also extends the collaboration between Microsoft and Anthropic from the May Series H investment down into the underlying compute layer.
Regional availability: eight data centers across North America, Europe, and Asia-Pacific
The initial available regions cover three major regions—North America (East US, West US, Canada Central), Europe (West Europe, North Europe), and Asia-Pacific (Australia East, Japan East)—for a total of eight Azure data centers.
Claude’s GA on Microsoft Foundry, and Anthropic’s previous collaborations with AWS Bedrock and Google Cloud Vertex AI, are proceeding in parallel; on April 7, another report said that Anthropic partnered with Broadcom and Google to obtain 3.5 gigawatts of TPU compute power, reflecting a landscape where these three hyperscalers jointly support Claude training and inference.
This article “Claude moves into Azure Foundry GA: first run of GB300 clusters” first appeared on .
On June 29, Anthropic will officially make the Claude model family generally available (GA) on Microsoft Foundry on Azure. With the Opus 4.8 and Haiku 4.5 dual models, Azure customers can use them directly in Azure clients, with support for Azure Active Directory authentication, unified billing, and full integration with the Anthropic Messages API. This deployment is among the first frontier AI services running on NVIDIA GB300 Blackwell Ultra GPU clusters in Azure. With the launch of both Opus 4.8 and Haiku 4.5, pricing is the same as buying directly from Azure: Opus 4.8 input pricing is USD 15 per million tokens, and output pricing is USD 30 per million tokens; Haiku 4.5 input pricing is USD 0.25 per million tokens, and output pricing is USD 1.25 per million tokens. Azure customers can call the service through the Foundry unified interface without needing to apply for an Anthropic account or handle cross-platform billing. Azure AD authentication allows enterprise customers to use their existing identity management mechanisms to control access permissions for using Claude models.
First NVIDIA GB300 Blackwell Ultra deployment: Azure Accelerator upgrade
What’s special about this deployment is the underlying infrastructure: it runs entirely on NVIDIA’s next-generation GB300 Blackwell Ultra GPU clusters and is the first frontier AI service on Azure to enable this architecture. GB300 is NVIDIA’s latest AI accelerator announced in June; compared with the previous generation GB200, it delivers significant improvements in inference throughput and energy efficiency. This deployment also extends the collaboration between Microsoft and Anthropic from the May Series H investment down into the underlying compute layer.
Regional availability: eight data centers across North America, Europe, and Asia-Pacific
The initial available regions cover three major regions—North America (East US, West US, Canada Central), Europe (West Europe, North Europe), and Asia-Pacific (Australia East, Japan East)—for a total of eight Azure data centers.
Claude’s GA on Microsoft Foundry, and Anthropic’s previous collaborations with AWS Bedrock and Google Cloud Vertex AI, are proceeding in parallel; on April 7, another report said that Anthropic partnered with Broadcom and Google to obtain 3.5 gigawatts of TPU compute power, reflecting a landscape where these three hyperscalers jointly support Claude training and inference.
This article “Claude moves into Azure Foundry GA: first run of GB300 clusters” first appeared on .